Professional Documents
Culture Documents
Control of Corporate Decisions
Control of Corporate Decisions
Management*
By
Milton Harris†
University of Chicago
And
Artur Raviv‡
Northwestern University
http://ssrn.com/abstract=990509
Milton Harris†
University of Chicago
and
Artur Raviv‡
Northwestern University
Activist shareholders have lately been attempting to assert themselves in a struggle with
management and regulators over control of corporate decisions. These efforts have met with mixed
success. Meanwhile, shareholders have been pressing for changes in the rules governing access to the
corporate proxy process, especially in regard to nominating directors. The key issue which these events
have brought to light is whether, in fact, shareholders will be better off with enhanced control over
corporate decisions. Proponents of increased shareholder participation argue that such participation is
needed to counter the agency problems associated with management decisions. In this view, boards of
directors do not exercise sufficient control over self-interested managers because management insiders
typically hand-pick directors through their control of the proxy process. Opponents offer several
arguments such as that shareholders lack the requisite knowledge and expertise to make effective
decisions or that shareholders may have incentives to make value-reducing decisions. In this paper, we
investigate what determines the optimality of shareholder control, taking account of some of the above
arguments, both pro and con. Our main contribution is to use formal modeling to uncover some factors
overlooked in these arguments. For example, we show that the claims that shareholders should not have
control over important decisions because they lack sufficient information to make an informed decision or
because they have a non-value-maximizing agenda are flawed. On the other hand, it has been argued that,
since shareholders have the “correct” objective (value maximization) and can always delegate the
decision to insiders when they believe insiders will make a better decision, shareholders should control all
major decisions. We show that this argument is also flawed.
JEL Classification Codes: G3, G34, G38
Keywords: corporate governance, shareholder democracy, direct shareholder participation, proxy process.
*
The authors are grateful for financial support from the Center for Research in Security Prices at the
University of Chicago Graduate School of Business.
†
Professor Harris, the corresponding author, is the Chicago Board of Trade Professor of Finance and
Economics at the Graduate School of Business, University of Chicago. He may be contacted at milt@uchicago.edu,
+1 (773) 702-2549, Graduate School of Business, University of Chicago, 5807 South Woodlawn Avenue, Chicago,
IL 60637, USA.
‡
Professor Raviv is the Alan E. Peterson Distinguished Professor of Finance at the Kellogg Graduate
School of Management, Northwestern University.
1
The leading proponent, at least in the academy, is Lucian Bebchuk. See Bebchuk (2005).
2
See, for examples, Bainbridge (2006), Lipton (2002), Stout (2006), and Strine (2006).
3
For example, Bebchuk (2005, pp. 881-882), arguing that shareholder ignorance is no excuse for denying
shareholders control, states, “After balancing the considerations for and against deference [to management], rational
shareholders might often conclude that deference would be best on an expected-value basis. Other times, however,
they might reach the opposite conclusion. Although shareholders cannot be expected to get it right in every case, it is
their money that is on the line, and they thus naturally have incentives to reach decisions that would best serve their
interests.”
4
Obviously, if the interests of shareholders are perfectly represented by independent directors on the board
whose information includes any private information of shareholders, then the issue becomes who should control the
board or various decisions made by the board, not whether shareholders should directly control these decisions.
This is the topic addressed in Harris and Raviv (2007a). Here, we make the opposite assumption that the board does
( s − ( a + p ) )
2
. (1)
For most of the paper, we assume that shareholders seek to maximize expected firm value, or,
equivalently, to minimize the expected loss derived from (1), given whatever information they have.7 It
follows that shareholders’ optimal decision is given by s = E ( a + p ) , where the expectation is
conditional on whatever information shareholders have about a and p .
Insiders, on the other hand, are biased relative to value maximization. In particular, we assume
that insiders seek to minimize, given whatever information they have, the expectation of the loss function
( s − ( a + p + b ) )
2
. (2)
where insiders’ bias, b, is a positive parameter that measures the extent of the agency problem between
shareholders and management.8 Insiders’ optimal decision is given by s = E ( a + p ) + b , where the
expectation is conditional on whatever information insiders have about a and p .
Because of the quadratic loss functions in (1) and (2), the difference between the expected loss
that results when the insiders make the decision and the expected-loss-minimizing decision, for any given
information, is b2. We therefore refer to b2 (and sometimes b) as the agency cost.
The shareholders’ information cannot be obtained by insiders, but may be communicated to them
by shareholders. Similarly, shareholders cannot obtain the insiders’ information directly, but insiders
not effectively represent the interests of shareholders. For evidence on the extent to which CEO involvement in the
selection of new board members, results in appointments of less independent directors, see Shivadasani and
Yermack (1999).
5
We will use the word “optimal” to mean value-maximizing throughout the paper, even in cases where
some shareholders have other objectives. We also assume that maximizing share value and maximizing firm value
are equivalent.
6
The specific assumptions that the optimal decision is the sum of a and p and that the loss is quadratic
can be generalized somewhat for some of our results, but these assumptions greatly simplify the analysis. We
believe that the insights derived from this model do not depend on these assumptions.
7
In section 3, we consider non-value-maximizing behavior on the part of shareholders.
8
It is not essential that b be positive. This is assumed only for expositional convenience.
In some cases, it will be more convenient to work with the standard deviations of the random
variables a and p , as well as those of other uniformly distributed random variables, instead of the
parameters A and P. Consequently, for any x ≥ 0, denote by σ ( x ) the standard deviation of a random
variable uniformly distributed on an interval of width x, i.e., σ ( x ) = x 12 . We will use σ a to denote
σ ( A) and σ p to denote σ ( P ) .
Because of the quadratic cost function in (1), it turns out that if an unbiased decision-maker
chooses s, the difference in firm value between knowing p (respectively, a ) and having no information
about p (respectively, a ) is exactly σ p2 (respectively, σ a2 ). We will therefore refer to σ p2 and σ p ( σ a2
and σ a ) as the importance of the shareholders’ (insiders’) information. We focus on the case in which
the agency problem (as measured by b) is severe relative to the importance of shareholders’ information.10
Formally, we assume
Assumption 2: σ p ≤ b .
Recall that control of a decision empowers the controlling party to make the decision or to
delegate it to the other party. If the controlling party does not delegate, it will make the decision based on
its own information and any information communicated to it by the other party.11 The sequence of events
is assumed to be the following. After observing its private information, the controlling party decides
whether to delegate to the other party or not. The party not making the decision may communicate some
or all of its private information to the decision maker. Finally, the decision maker chooses s and firm
value is realized.
2 Base Case Analysis
In this section, we analyze the case in which shareholders are value-maximizers and understand
perfectly the extent of their private information as well as all other parameters of the model. Essentially,
the current model is a special case of the model in Harris and Raviv (2007a).12 Consequently, we will
9
Our results depend only on the widths of the supports of a and p , not on their locations.
10
It will be shown below that this assumption guarantees that management will not want to delegate to
shareholders and that shareholders will never reveal any of their private information to management.
11
We do not consider the possibility of constrained delegation, i.e., that one party allows the other party to
choose s subject to a constraint. One can show, for example, that if the shareholders have no private information, it
is optimal for them, when they delegate to insiders, to put an upper bound on insiders’ choice that is sometimes
binding. When shareholders have private information, the problem is more complicated, since their choice of a
constraint may convey information. Since this model is sufficiently complicated already, we leave this possibility
for future work.
12
Shareholders in the current model are like the outsiders in Harris and Raviv (2007a), while
management/insiders in the current model correspond to insiders in Harris and Raviv (2007a). In the current model,
we drop the assumption that outsiders (shareholders) must pay a cost to acquire their private information.
s ( p, r ) = a ( r ) + p , (3)
where a ( r ) = E ( a r ) is the mean of shareholders’ posterior belief about a , given the insiders’ report, r.
Because of the agency problem, insiders will not fully reveal their information. More precisely,
in the Pareto-best Bayes equilibrium of the game in which shareholders choose s, insiders will partition
the support of a , [ 0, A] , into cells [ ai , ai +1 ] (of unequal widths) and report a value that is uniformly
distributed on the cell in which the true realization of a lies. 13 Thus shareholders learn only the cell in
which the true value of insiders’ information lies, and their posterior belief is that a is uniformly
a + ai +1
distributed on that cell. It follows that if the report r is in [ ai , ai +1 ] , a ( r ) = i . The number of cells
2
in the partition is denoted by N ( b, A ) (see Harris and Raviv (2007b) for an explicit formula). Note that
the number of cells is a measure of the extent to which insiders communicate their information to
shareholders. For example, if there is only one cell, N ( b, A ) = 1 , insiders communicate nothing
shareholders do not already know. This will be the case, for example, if insiders’ information is less
important than agency cost, i.e., σ a ≤ b . On the other hand as N ( b, A ) gets very large (as will be the
case if agency cost, b, approaches zero), the information communicated approaches perfect information
about a . We state the following result about the function N ( b, x ) for future reference (the proof can be
found in Harris and Raviv (2005)).
Lemma 1. N ( b, x ) = 1 if and only if x ≤ 4b or, equivalently, σ ( x ) ≤ 2b 3.
Since insiders do not fully communicate their private information, there is a consequent loss in
firm value. Let L ( b, A ) denote this information cost, i.e., the expected loss in firm value due to having
only information about a that is transmitted in equilibrium (as opposed to full information). That is,
L ( b, A ) = E ⎡⎣ a ( r ( a ) ) + p − ( a + p ) ⎤⎦ = E ⎡⎣ a ( r ( a ) ) − a ⎤⎦ .
2 2
(4)
13
The game in which shareholders choose s is analyzed formally in Harris and Raviv (2005). Here we
simply summarize the results and provide intuition.
where p̂ = E ( p delegation ) is the mean of the insiders’ posterior belief about p given the fact that the
decision has been delegated. Therefore, to analyze the equilibrium in this case, one must first understand
in which circumstances, i.e., for which values of p , shareholders will delegate.
To understand when shareholders will delegate, consider the loss in firm value from delegating
and the loss from not delegating. There are two components of the loss from delegating, namely the
direct agency cost, b 2 , and the loss due to imperfect communication of shareholders’ private information.
The loss from not delegating is the loss due to imperfect communication of insiders’ private information,
i.e., L ( b, A ) .
We argue that the equilibrium delegation region is of the form [ p*, P ] for some threshold
p* ∈ [ 0, P ] , i.e., that shareholders will delegate if and only if p exceeds some threshold p * . To see this,
14
An explicit formula for L is given in Harris and Raviv (2007b) where it is shown that this expected loss
depends only on the agency cost, b, and the width A of the support of a .
p̂ is the midpoint of D, p̂ + b is to the right of the midpoint of D. If d 2 < P , then for some values of p
strictly between d 2 and P, p is closer to p̂ + b than some values of p in the delegation region D. But this
means that there are values of p not in the delegation region for which shareholders would prefer to
delegate. Hence, if d 2 < P , D cannot be an equilibrium delegation region, i.e., any equilibrium
delegation region must have P as its right endpoint. This shows that delegation by shareholders, if it
occurs, will occur for all values of p above some threshold, p * .
It is convenient to denote by d the width of the interval of values of p over which shareholders
delegate, i.e., d = P − p * . We characterize p * and d more precisely in the following proposition.16
Proposition 1: If shareholders are in control of a decision, they will delegate the decision if and
only if the realized value of their private information exceeds a threshold p* ∈ [ 0, P ] satisfying
the following three conditions:
If σ a ≤ b , shareholders never delegate, i.e., p* = P and d = 0 ; (6)
LS =
P
(
d 2 2
)⎛ d⎞
b + σ ( d ) + ⎜ 1 − ⎟ L ( b, A ) .
⎝ P⎠
(9)
To understand this expression, first recall that if shareholders delegate, they do not share any information
about p with insiders other than the fact that p ≥ p * , due to Assumption 2. Consequently, the expected
loss in firm value if shareholders delegate is the agency cost, b 2 , plus the loss from knowing only that
p ∈ [ p*, P ] , σ ( P − p *) = σ ( d ) . Thus the first term on the right hand side of (9) is the probability that
2 2
shareholders delegate, d P , times the expected loss if they do, b 2 + σ ( d ) . The second term on the right
2
15
This is an intuitive argument that assumes the delegation region is an interval. See Harris and Raviv
(2005) for a formal proof that there is an equilibrium in which shareholders delegate if and only if p exceeds a
threshold.
16
As mentioned above, if shareholders have no private information ( σ p = 0 ), they will delegate if and only
if insiders’ information is more important than agency costs ( σ a ≥ b ). This is a special case of conditions (6)−(8) .
⎡d
( ⎛ d⎞
) ⎤
Δ ≡ LM − LS = b 2 + σ p2 − ⎢ b 2 + σ ( d ) + ⎜ 1 − ⎟ L ( b, A ) ⎥ ≥ 0 .
⎣P
2
⎝ P⎠ ⎦
(11)
The main result of this section is stated in the following proposition (proved in the appendix).
Proposition 2.
(i) If the information of insiders is less important than agency costs ( σ a < b ), then for any
σ p ∈ [ 0, b ] , shareholder control is optimal ( Δ > 0 ).
(ii) For every σ a ≥ b , there are two boundaries for σ p , σ pL (σ a ) and σ Up (σ a ) , such that when
the importance of the shareholders’ information σ p ∈ (σ pL (σ a ) ,σ Up (σ a ) ) , insider control is
strictly optimal ( Δ < 0 ). When the importance of the shareholders’ information
σ p > σ Up (σ a ) , shareholder control is strictly optimal ( Δ > 0 ). When σ p ≤ σ pL (σ a ) or
σ p = σ Up (σ a ) , control is irrelevant ( Δ = 0 ).
Proposition 2 shows that the parameter space (σ a , σ p ) can be divided into three regions as shown
in Figure 1. It is optimal for insiders to control decisions for which the importance of the information of
the two parties lies between the upward sloping curves defined by σ p = σ Up (σ a ) and σ p = σ pL (σ a ) . For
decisions for which management’s information is less important than agency cost ( σ a < b ) or for which
p
Im p o r t a n c e o f S h a r e h o ld e r s ’ In f o r m a t io n , σ 7
U
σp
6 S h a r e h o ld e r C o n tr o l
5 is O p tim a l
L
4 σp
0
In s id e r C o n tr o l
3 45
is O p tim a l
2
1 S h a r e h o ld e r
C o n tr o l = In s id e r C o n t r o l
0
0 2 4 6 8 10 12 14 16 18
Im p o r t a n c e o f In s id e r s ’ In f o r m a t io n , σ a
Figure 1
This graph shows how optimal control varies with the two information parameters, σ a ,
the importance of insiders’ information, and σ p , the importance of shareholders’
information. For this figure, b = 8.
Given that shareholders’ objective is to maximize firm value and given that, if shareholders
believe that insiders’ choice of s will result in higher value than their own, they may delegate the decision
to insiders, one might ask why it would ever be optimal for insiders to control the decision. Indeed, from
the point of view of assigning control, which is done ex ante, if shareholders were able, ex ante, to
commit to a delegation threshold, it would always be optimal to assign them control.
For suppose shareholders could announce and commit to any delegation cutoff, P − x with
x ∈ [ 0, P ] , before observing their private information. Since shareholders are committed to delegate if
and only if p ≥ P − x , if shareholders delegate, insiders correctly infer that p ≥ P − x . In this case, the ex
ante expected loss from shareholder control is given by equation (9) with d replaced by x:
x 2
P
( ⎛
)
x⎞
b + σ ( x ) + ⎜ 1 − ⎟ L ( b, A ) .
2
⎝ P⎠
17
Note that we are not claiming the solution to this problem is an optimal delegation policy. We claim only
that, among policies that are characterized by a lower threshold for delegating, the solution is optimal. Nevertheless,
we show that, even if one is restricted to such threshold policies, shareholder control is always optimal.
18
s * + a is insiders’ choice of s when the threshold is p * , but since a cancels in calculating shareholders’
expected loss from delegating, s * is the relevant comparison point.
L(b,A)
Threshold = p*
Threshold = p ′
0
0 p ′ p0 p* s* P
S h a r e h o ld e r s ’ In f o r m a t io n , p
Figure 2
This figure shows the loss to shareholders from delegating as a function of the realization
of their private information. The loss is shown for two values of the delegation threshold,
the equilibrium value, p * (dark blue or black curve), and a lower value p′ < p * (pink or
gray curve). The value of p that minimizes shareholders’ expected loss from delegating,
given that insiders believe shareholders delegate for p ≥ p * , is shown as s * . The red
(or dark) triangular region on the left represents the extra expected cost of reducing the
threshold from p * to p ′ due to having to delegate for some values of p for which, ex
post, not delegating is less costly. The green (or gray) region minus the small red (or
dark) region on the right represents the expected benefit of reducing the insiders’ choice
of s by delegating for smaller realizations of p . For this graph, b = 10 , σ p = 9 , and
σ a = 50 .
Proposition 2 has several interesting and important implications. First, when shareholders have
no private information ( σ p = 0 ), it is strictly optimal for them to be in control when management’s
information is less important than agency cost ( σ a < b ) and weakly optimal when management’s
information is more important than agency cost ( σ a ≥ b ). This is because, when shareholders have no
private information, they make an ex-ante-optimal delegation decision, namely to delegate if
management’s information is more important than agency cost and not otherwise. On the other hand,
when shareholders do have private information, it is sometimes optimal for them not to be in control as
shown in Proposition 2. Thus we obtain the somewhat counterintuitive result that shareholders should
19
See, e.g., Bainbridge (2006).
E ( p + a + b − ( p + a ) ) = σ 2p + b 2 .20
2
(12)
Since shareholders always delegate in this case, their ex ante expected loss consists of the loss due to
having no information about p and the agency cost.
Now suppose that q < p * , so that shareholders do not delegate, and insiders observe a = a .
From the point of view of insiders, shareholders are biased for two reasons. First, shareholders are biased
by −b , i.e., toward smaller s relative to insiders’ optimum. Second, shareholders are also biased by their
misperception of p . This misperception bias is given by e = q − p . Since insiders do not observe p ,
they are interested in shareholders’ average misperception bias, e = E ( e ) = q − p . Therefore, when
shareholders misperceive p and do not delegate, insiders view them as biased by −B , where
− B = −b + e . (13)
We refer to −B as shareholders’ effective bias.
In the equilibrium of the communication game when shareholders do not delegate, insiders will
behave exactly as in section 2, except that they will replace −b by the shareholders’ effective bias −B .
Shareholders’ average misperception bias may offset or reinforce the original bias, depending on whether
shareholders’ misperception bias is positive or negative on average. If the shareholders perceive p
correctly on average, i.e., e = 0 , then B = b , and the communication game in this case is the same as in
section 2.1. If shareholders’ perception of p is larger than its average value ( e > 0 ), this reduces their
bias toward smaller s from the point of view of insiders and, interestingly, improves communication from
insiders to shareholders. Just the opposite is true if shareholders’ perception of p is smaller than its
average value.21
20
Note that this is not how shareholders perceive their expected loss. We are interested only in their actual
expected loss, since it is this loss that determines whether it is optimal for shareholders to control the decision.
21
A formal proof of the argument in this paragraph is given in the appendix. Note that the maximum
number of elements in the equilibrium partition of [ 0, A] is now N ( B, A ) instead of N ( b, A ) . Since B may be
negative, to allow for negative biases, throughout the remainder of this paper the functions N ( x, y ) and f ( x, y )
will be redefined as the original functions evaluated at ( x , y) . There is no need to redefine L ( x, y ) , since L
2
depends on x only through x .
This expression differs from shareholders’ expected loss when they do not delegate in the base case of
section 2.1, L ( b, A ) , in three respects. First, the loss due to imperfect communication from insiders
about a , L ( B, A ) , reflects the shareholders’ effective bias (from the point of view of insiders) rather than
the actual bias. Second, there is an additional loss of σ p2 due to the fact that shareholders do not observe
p . Third, there is an additional loss of e 2 due to the fact that shareholders misperceive p .
To summarize, if shareholders are in control, their actual expected loss is given by (12) if they
always delegate ( q ≥ p * ) and is given by (14) if they never delegate ( q < p * ).
22
The function r ( a ) is the equilibrium reporting function from section 2.1, except that b is replaced by B.
The equality follows from the fact that E ⎡⎣ a ( r ( a ) ) ⎤⎦ = E ( a ) and the independence of a and p .
23
Formally, insiders’ expected loss from delegating is the loss due to imperfect communication of insiders’
information plus Var ( e ) + B 2 . Since Var ( e ) = σ p2 , insiders’ expected loss from delegating is at least σ p2 and
greater than that if B ≠ 0 . The insiders’ loss from not delegating is simply σ p2 . Consequently, insiders will never
delegate in this situation. Interestingly, if shareholders have some private information, insiders may strictly prefer to
delegate. The reason for this is that shareholders’ misperception may make their effective bias very small resulting
in communication from insiders about a that is very precise. In this case, by delegating, insiders can take
advantage of shareholders’ information while losing little due to shareholders’ effective bias and imprecise
communication about a . As an extreme example, suppose shareholders observe α p + (1 − α ) p + b (but believe
they observe p ), where α ∈ ( 0,1) and b < (1 − α ) p (the last condition guarantees that shareholders’ observation is
in [ 0, P ] ). In this case, e = (1 − α )( p − p ) + b , e = b , and B = 0 . Consequently, by delegating, insiders lose
nothing from imprecise communication of their private information, nor from shareholders’ effective bias, i.e., their
expected loss is Var ( e ) = (1 − α ) σ p2 . Since their expected loss from not delegating is σ p2 , we see that insiders
2
Now consider the case in which σ a < b . In this case, the equilibrium and optimal delegation
decision for shareholders from section 2.1 is not to delegate regardless of the realization of p , i.e.,
p* = P . Consequently, q < p * , so shareholders also prefer not to delegate in the present situation in
which they are misinformed, and the optimality of shareholder control is determined by the above
comparison. The loss due to imperfect information about a cannot exceed the loss from knowing
nothing about a , σ a2 , however. Thus the relevant part of the expected loss if shareholders are in control
in this case is not greater than e 2 + σ a2 . Since σ a < b , if e is sufficiently small, it is optimal for
shareholders to be in control, even though they are misinformed and never delegate.24 Essentially, in this
24
For example, if b = 1 , e = 0.7575 (so B = 0.2425 ), and σ a = 0.28141147 < b ,
e 2 + L ( B, A ) = 0.6524106 < 1 = b 2 .
shareholders’ optimal decision is the same as that of insiders’, except that b is replaced by β . That is,
NVM shareholders will choose s = E ( a + p ) + β , where the expectation is conditional on whatever
information they have about a and p . Here, the parameter β measures the extent to which these
shareholders will deviate from the optimal decision to further their social agenda.25 Note that β could be
25
We use the term “optimal” in this section to mean firm-value maximizing.
similar role to b − e in the previous section, we redefine B = b − β and refer to it as the net bias. We
also assume that the parameter values are such that insiders, if in control, never delegate to shareholders.26
Note that B may be positive or negative.
If B > 0 , all the results of section 2 apply, except that B replaces b in all calculations. In
particular, p * and d = P − p * satisfy (6)−(8) with b replaced by B.
If B < 0 , then NVM shareholders delegate when p ∈ [ 0, p *] , and p * and d are calculated as in
section 2.1 except that b is replaced by B and p* = d .27
Also, for B < 0 , insiders never delegate to shareholders if and only if σ p ≤ B . Thus the
counterpart of Assumption 2 in this case is σ p ≤ B , which also implies that (NVM) shareholders do not
communicate any private information to insiders (other than what may be communicated by their
delegation decision).
The expression in (15) is the same as in the base case (equation (9)), except for three effects that are
similar to those discussed in section 3.3. First, the size of the delegation region, d, is determined by the
net bias B rather than insiders’ bias b. Second, the loss when shareholders do not delegate is increased by
the cost of the NVM shareholders’ bias, β 2 . This effect obviously reduces the attractiveness of
shareholder control and is the effect on which opponents of shareholder control focus. Third, the loss due
to imperfect communication from insiders is determined by the net bias instead of insiders’ bias. Since
the net bias can be smaller than the insiders’ bias, communication of insiders’ information can be more
precise than in the base case, resulting in smaller loss. As is shown in Proposition 5 below, this effect
causes shareholder control to be optimal in some cases.
26
Exactly what this requires will be made explicit below.
27
The argument for delegation when p is below a threshold if B < 0 is essentially the same as the
argument for delegation when p is above a threshold in section 2.1, except that, when B < 0 , insiders’ choice of s
results in shareholders’ expected loss from delegation being minimized at a value of p that is below the midpoint of
the delegation region, instead of above it.
(
d 2
P
2
)
⎛ d⎞
b + σ ( d ) + ⎜ 1 − ⎟ ( β 2 + L ( B, A ) ) ≤ b 2 + σ p2 .
⎝ P⎠
(16)
Note that if σ a ≤ B , neither party will delegate to the other nor will they communicate any of
their private information. Therefore, control is equivalent to making the decision with no information
from the other party. The expected loss in firm value resulting from control by a given party is simply
that party’s agency cost plus the loss from not observing the other party’s information. Condition (16)
becomes
β 2 + σ a2 ≤ b 2 + σ p2 or β 2 − b 2 ≤ σ p2 − σ a2 .
The second inequality in the above condition makes it clear that optimality of shareholder control
in this case hinges on a comparison between the difference in the importance of the parties’ information
and the difference in their agency costs. It will be convenient to restate the condition in terms of the net
bias, B:
σ a2 ≤ σ p2 − ( B 2 − 2bB ) . (17)
The main result of this section (proved in the appendix) is to characterize optimal control of
decisions for various values of the net bias B and the importance of insiders’ information σ a2 , for fixed
values of insiders’ bias b and the importance of shareholders’ information σ p .
• for decisions for which B < −σ p , insider-control is strictly optimal if σ a2 < G ( B ) , and
otherwise insider-control and outsider-control are equivalent;
• for decisions for which σ p < B < b − p , shareholder-control is strictly optimal if
σ a2 < G ( B ) , and otherwise, insider-control and outsider-control are equivalent;
• for decisions for which b − p < B < B0 , shareholder-control is strictly optimal if
σ a2 < H ( B ) , insider-control is strictly optimal if H ( B ) < σ a2 < G ( B ) , and otherwise,
insider-control and outsider-control are equivalent;
• for decisions for which B0 < B , insider-control is strictly optimal if σ a2 < G ( B ) , and
otherwise, insider-control and outsider-control are equivalent.
The proposition is depicted in Figure 3.
28
This is the richest case. When this inequality fails, the result is qualitatively similar.
100
G(B)
B2 B2
50
Insider Control Insider Control
is Optimal is Optimal
H(B)
Share-
holder
Control is Optimal B0
0
-1 0 −σ p 0 σp b − p b − β0 2b 10
N e t B ia s o f N V M S h a r e h o ld e r s , B
Figure 3
This figure shows, for various combinations of values of the net bias, B, and the
importance of insiders’ information, σ a2 , which party optimally controls the decision.
The values of other parameters are b = 4 , σ p = 1 . These values imply that p = 3 and
B0 = 4 + 17 ≈ 8.123 .
For decisions for which the net bias is positive, NVM shareholders are either less biased in the
same direction as insiders ( 0 < β < b ) or are biased in the opposite direction ( β < 0 ). In this case, there is
a tradeoff. When the net bias, B = b − β , is small, but still more important than shareholders’ information
( B > σ p ), NVM shareholders’ bias, β , is similar to that of insiders but smaller ( 0 < β < b ). Since the net
bias is small, insiders are willing to communicate much of their information to shareholders if
shareholders are in control and decide not to delegate. Since NVM shareholders’ bias is smaller than that
29
Actually, it is shown in Proposition 5 that H is downward sloping for B ≥ b 2 .
min
x∈[ 0, P ]
x
P
( 2
)
⎛
⎝ P⎠
x⎞
σ ( x ) + b 2 + ⎜ 1 − ⎟ L ( b, A ) , (18)
x2
g ( b, x ) ≡ + b2 . (19)
4
Then the derivative of the objective function in (18) with respect to x is
g ( b, x ) − L ( b, A )
. (20)
P
Clearly, g is increasing in its second argument, so the objective function in (18) is convex. Also,
for any x ∈ ( 0, P ) ,
2
⎛x ⎞
f ( b, x ) = ⎜ + b ⎟ > g ( b, x ) . (21)
⎝2 ⎠
Since p* ∈ ( 0, P ) , so d = P − p* ∈ ( 0, P ) , from (7) and (21), we have
f ( b, d ) = L ( b, A ) > g ( b, d ) .
Therefore, from (20), the objective function in (18) is strictly decreasing in x at x = d . Since d ∈ ( 0, P ) ,
we must have x* > d . Consequently, p ** = P − x* < P − d = p * . Q.E.D.
30
Note that we do not claim that delegating when p ∈ [ p **, P ] is an optimal ex ante delegation policy for
shareholders. We claim only that this policy is optimal in the class of policies in which shareholders delegate when
p is above some cutoff value and that it is better than the equilibrium ex post delegation policy of delegating when
p ∈ [ p*, P ] .
d = 2 ⎡ L ( b, A ) − b ⎤ . (27)
⎣ ⎦
Clearly, the formula for d in (27) is valid only if it results in a value between 0 and P. If
L ( b, A ) ≤ b 2 , then (27) results in d ≤ 0 . In this case, insiders do not have sufficient information to
warrant delegating to them regardless of the realization of p , so d = 0 and p* = P . As mentioned
above, however, L ( b, A ) ≤ b 2 if and only if σ a ≤ b , so we have that d = 0 and p* = P if and only if
σ a ≤ b . If L ( b, A ) ≥ f ( b, P ) , or, using (25), P ≤ 2 ⎡ L ( b, A) − b ⎤ , then (27) results in d ≥ P . In this
⎣ ⎦
case, insiders have sufficient information to warrant delegating to them regardless of the realization of p ,
so d = P and p* = 0 . Q.E.D.
Proof. First suppose σ a < b . In this case, as shown in Proposition 1, d = 0 , so from (11),
Δ = b 2 + σ p2 − L ( b, A) > σ p2 > 0 ,
since, from Lemma 1 of Harris and Raviv (2007b), σ a < b implies that L ( b, A ) < b 2 . This proves part (i).
Henceforth, we assume σ a ≥ b .
Define σ pL by
2
⎛ σ pL 12 ⎞
(
f b, σ pL 12 = ⎜)
⎜ 2
+ b ⎟ = L b, σ a 12 ,
⎟ ( ) (28)
⎝ ⎠
or
σ pL (σ a ) =
2 ⎛
( ⎞
⎜ L b, σ a 12 − b ⎟ .
12 ⎝ ⎠
) (29)
From Lemma 1 of Harris and Raviv (2007b), σ a ≥ b implies that σ pL as given in (29) is non-negative.
( )
Since σ a ≤ b implies that L ( b, A ) = L b, σ a 12 = σ a2 , it is obvious from (29) that σ pL ( b ) = 0 .
Since L is strictly increasing in its second argument (Lemma 1 of Harris and Raviv (2007b)), it is also
obvious from (29) that σ pL (σ a ) is increasing in σ a , ∀σ a > b , as claimed in part (iii).
The next step is to develop a formula for σ Up (σ a ) . Assuming for the time being that
{ }
σ Up (σ a ) > σ pL (σ a ) and using d = min σ pL (σ a ) 12, P , we can write the condition defining σ Up (σ a ) as
( )
σ Up R − (σ Up ) = σ pL R − (σ pL ) ,
2
( 2
) (30)
( )
where R = L b, σ a 12 − b 2 . Rewrite (30) as
(σ ) + σ Up σ pL + (σ pL ) − R = 0 .
U 2 2
p
−σ pL + 4 R − 3 (σ pL )
2
σ Up = . (32)
2
For σ Up to be given by equation (32), we need only show that this value exceeds σ pL . For this, it
suffices to show that R > 3 (σ pL ) . But from (29) and the definition of R, we have that R > 3 (σ pL ) if and
2 2
only if ( ) ( )
L b, σ a 12 + b > L b, σ a 12 − b , which is clearly true, since b > 0 . Consequently, we have
shown that σ Up is indeed given by equation (32) and that σ Up (σ a ) > σ pL (σ a ) , as claimed in (iii). For
σ a = b , R = σ a2 − b 2 = 0 , and, as shown previously, σ pL = 0 . Consequently, σ Up ( b ) = 0 , as claimed in
(iii).
To show that σ Up (σ a ) < σ a , from (32), it suffices to show that
( )
R = L b, σ a 12 − b 2 < σ a2 + σ aσ pL + (σ pL ) .
2
(33)
( )
It is easy to check that L b, σ a 12 ≤ σ a2 .31 Therefore (33) is clearly satisfied since b, σ a , and σ pL are all
positive.
To complete the proof of part (iii), it remains to show that σ Up (σ a ) is increasing in σ a . For this,
( )
z = L b, σ a 12 . Then, substituting for R and σ pL in (32), we have
1⎡ 2 ⎤
σ Up = ⎢ − ( z − b ) + 3 ( z 2 − b 2 ) + 2b ( z − b ) ⎥ . (34)
2 ⎣ 12 ⎦
It is easy to check that the derivative of the right hand side of (34) with respect to z is positive if and only
if 3z 2 + 2bz + b 2 > 0 , which is clearly true. This completes the proof of part (iii).
31
( ) ( )
If N b, σ a 12 = 1 , then L b, σ a 12 = σ a2 , and we are done. Suppose N b, σ a 12 = n ≥ 2 . Then ( )
σ a2 b ( n − 1)
2 2
( bn )
2
(
L b, σ a 12 = ) n 2
+
3
< σ a2 if and only if σ a2 >
3
. But, as shown in Lemma 1 of Harris and Raviv
( bn ( n − 1) )
2
( bn )
2
( )
(2007b), N b, σ a 12 = n implies that A > 2bn ( n − 1) or σ a2 >
3
≥
3
, for n ≥ 2.
σ pL σ pU
Net Gain to Shareholder Control
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
-1
-2
-3
-4
-5
Figure 4
This graph shows the net gain to shareholder control, Δ , as a function of the importance
of shareholders’ information. Shareholder control is optimal whenever the importance
of shareholders’ information, σ p , exceeds σ Up ≈ 3.45 . For this figure, b = 4, σ a = 8 ,
and σ pL = 0.95658525 < b.
⎡⎛ a + a ⎞ ⎤ ⎛a +a
2
⎞
2
E ⎢⎜ i −1 i + e − a − b ⎟ ⎥ = ⎜ i −1 i − a − ( b − e ) ⎟ + σ p2 . (35)
⎢⎣⎝ 2 ⎠ ⎥⎦ ⎝ 2 ⎠
Proof. Optimal control is obtained by comparing LM with shareholders’ expected loss from
either (12) or (14). Comparison with (12) shows that, when q ≥ p * , control doesn’t matter. This proves
the last part of the proposition.
If q < p * , comparison of LM with (14) shows that it is optimal for shareholders to be in control
if and only if
b 2 > e 2 + L ( B, A ) . (36)
First consider the case in which σ a < b . In this case, p* = P (see Proposition 1). Consequently,
q < p * . Since σ a < b , we have L ( B, A ) ≤ σ a2 < b 2 , so if e is sufficiently small, it is optimal for
shareholders to be in control.32 This establishes the claim in item (i).
Now assume σ a ≥ b and q < p * . If shareholders are in control, they never delegate. Suppose
e ≤ 0 . Then, from (13), B ≥ b , and, using Lemma 1 of Harris and Raviv (2007b),
L ( B , A ) ≥ L ( b, A ) ≥ b 2 ,
32
For example, if b = 1 , e = 0.7575 (so B = 0.2425 ), and σ a = 0.28141147 < b ,
e 2 + L ( B, A ) = 0.6524106 < 1 = b 2 .
β0 =
1
2
(b − b 2 + 2σ 2p .) (37)
Proof. From (37), it is obvious that β 0 < 0 and that −b < β 0 if and only if b > σ p 2 . Clearly,
b − β 0 < 2b if and only if −b < β 0 . Q.E.D.
• for decisions for which B < −σ p , insider-control is strictly optimal if σ a2 < G ( B ) , and
otherwise, insider-control and outsider-control are equivalent;
• for decisions for which σ p < B < b − p , shareholder-control is strictly optimal if
σ a2 < G ( B ) , and otherwise, insider-control and outsider-control are equivalent;
• for decisions for which b − p < B < B0 , shareholder-control is strictly optimal if
σ a2 < H ( B ) , insider-control is strictly optimal if H ( B ) < σ a2 < G ( B ) , and otherwise,
insider-control and outsider-control are equivalent;
• for decisions for which B0 < B , insider-control is strictly optimal if σ a2 < G ( B ) , and
otherwise, insider-control and outsider-control are equivalent.
2
(
⎛P ⎞
)
Proof. Define G ( B ) as the value of σ a2 such that L B,σ a 12 = ⎜ + B ⎟ . It is easy to
⎝ 2 ⎠
( ) ( )
check, using the facts that L B,σ a 12 = σ a2 for σ a2 ≤ B 2 and L B,σ a 12 → ∞ as σ a → ∞ , and L is
continuous in its second argument (see Lemma 1 in Harris and Raviv (2007)), that such a value of σ a2
exists and is larger than B 2 . Thus G ( B ) > B 2 for all B. Since L depends on B only through B 2 , and
2
⎛P ⎞
⎜ + B ⎟ depends on B only through B , G is symmetric with respect to B = 0 . Finally, since L and f
⎝ 2 ⎠
are continuous in B (for B ≠ 0 ), so is G. From the definition of G, for σ a2 ≥ G ( B ) , d = P , i.e.,
shareholders, if in control, always delegate to insiders. In this case, control is irrelevant, since insiders
always make the decision with no information from shareholders, regardless of control, i.e., (16) is
satisfied as an equality. For the remainder of the proof, we consider only the case in which σ a2 < G ( B ) .
P ⎝ P⎠ P ⎝ P ⎠⎝ 2 ⎠
We claim that the function F is strictly concave in d on ( 0, P ) . To see this, note that
1 2⎛d ⎞ 1 2B
F ′′ ( d ) = − ⎜ + B ⎟< − .
2 P⎝ 2 ⎠ 2 P
But P = 12σ p ≤ 12 B < 4 B . Consequently, F ′′ ( d ) < 0 and F is strictly concave as claimed. Also
F ( 0 ) = B 2 + β 2 = b 2 + 2 B ( B − b ) , and F ( P ) = σ p2 + b 2 = right hand side of (16).
Next, suppose B > σ p . Then it is easy to check that F ( 0 ) < σ 2p + b 2 if and only if B < b − β 0 .
Consequently, if B ≥ b − β 0 , F ( 0 ) ≥ σ 2p + b 2 , so insider control is strictly optimal. Now consider
σ p < B < b − β 0 , so that F ( 0 ) < σ 2p + b 2 . There are two possible cases. Since F is strictly concave, if
F ′ ( P ) ≥ 0 , then F ( d ) < σ p2 + b 2 for all d ∈ ( 0, P ) which implies that it is strictly optimal for
shareholders to control regardless of the value of B 2 < σ a2 < G ( B ) . If F ′ ( P ) < 0 , then there exists a
unique d 0 ∈ ( 0, P ) such that F ( d ) ≤ σ p2 + b 2 for all d ≤ d 0 , F ( d ) > σ 2p + b 2 for all d > d 0 , and
F ′ ( d 0 ) > 0 . That is, it is optimal for shareholders to control if and only if d ≤ d 0 . But d is a continuous,
⎧⎪ H 0 ( B ) for B ∈ [b − p, b − β 0 ] ,
H ( B) = ⎨
⎪⎩ H1 ( B ) for B ∈ [b − β 0 , B0 ].
∂F ( d )
Consequently, > 0 for all d ∈ ( 0, P ) if B ≥ b 2 . Since F ′ ( d 0 ) > 0 , it follows that, for B ≥ b 2 ,
∂B
d 0 is decreasing in B. Therefore, so is H 0 ( B ) . Since we have already shown that H1 ( B ) is decreasing
in B for B ≥ b − β 0 , H ( B ) is decreasing in B for B ≥ b 2 .
Q.E.D.